OpenAI Agents SDK
App : A product you use directly
OpenAI's open source framework for building agents and agent teams: handoffs between agents, guardrails, sessions, and built-in tracing.
Simple and quick to get started with, ideal for agents that hand off to one another. Optimized for OpenAI models, usable with other providers via adapters.
Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.
What it does, concretely
Building an AI agent that delegates tasks to other agents, manages memory across turns, and validates its own outputs usually requires assembling several libraries and writing a lot of coordination code. OpenAI Agents SDK reduces this work to a few clear primitives: agents, handoffs, guardrails, and sessions.
In practice, you define an agent by giving it instructions and tools (Python functions, MCP servers, hosted tools), then launch a run with Runner. The SDK handles the turn loop, tool calls, handoffs to other agents, and input/output validation through guardrails. Four execution modes are available: classic text agent, sandbox agent (in an isolated workspace for file inspection or command execution), realtime agent (low-latency voice over WebSocket), and voice pipeline (speech-to-text, agent, text-to-speech).
The SDK is aimed at Python developers who want to build multi-agent workflows in production without learning a large number of new abstractions. Python 3.10 or newer is required, and the tool is used from the command line.
It is not suited for those looking for a graphical interface, a no-code tool, or who want to manage the model call loop and state handling themselves: in that case, OpenAI's Responses API is a better fit.
What it offers
- ·Text agents: define an agent with instructions, tools, and guardrails; the SDK manages the turn loop until the task is complete.
- ·Sandbox agents: run an agent in an isolated workspace to inspect files, execute commands, or apply patches on longer-running tasks.
- ·Realtime agents: build low-latency voice agents with gpt-realtime-2.1 over WebSocket, with automatic interruption detection.
- ·Voice pipeline: combine speech-to-text, an agent workflow, and text-to-speech into a single processing chain.
- ·Agent handoffs: an agent can delegate a subtask to another specialized agent, or use another agent as a tool.
- ·Guardrails: validate inputs and outputs in parallel with agent execution, with fast failure if a check does not pass.
- ·Sessions and memory: automatic conversation history management across turns, with support for Redis, SQLAlchemy, and other backends.
- ·Built-in tracing: visualize, debug, and monitor agent runs, with export to the OpenAI dashboard for evaluation and model fine-tuning.
In practice
- Price
- Free
- Commercial use
- Commercial use: yes
- Available on
- cli
- Licence
- MIT
- FR interface
- No
Security and data
- Rating
- Security: one caveat, see the card
- Your data
- Local or cloud depending on use
- Maintenance
- Active
- Last release
- 15 October 2026
Links and repository checked on 25 October 2026 · added on 27 September 2026
How to install OpenAI Agents SDK
pip install openai-agents